Source: github.com/cline/cline/releases/tag/cli-v3.0.75
One-shot runs, meaning those started with a prompt argument or piped stdin, now use yolo mode instead of act mode because no one is present to approve tools. Act mode still applies when auto-approve is off, when a mode flag is passed, when plan mode is saved, or for /team prompts. Yolo mode has no team tools, so users who need them should pass --act. The --timeout flag and Ctrl+C can now also stop the first turn of a one-shot run.
Several fixes target model behavior. Read results are no longer truncated to 8k characters and arrive in full up to the tool's limit of 2000 lines or about 48k characters. Fixes cover Reflection Beam failing on compaction or model switches, summarizer requests using reasoning_effort set to none, reasoning settings on custom Model Configuration overrides, and reasoning levels on the native OpenAI provider. The CLI also no longer reads .env or bunfig.toml from the launch directory.
The model catalog adds 94 providers from models.dev and drops Lilac and Tempr. Default models changed for 67 other providers, mostly to Kimi K3.1 Pro, Gemini 4 Neon or DeepSeek V4.1 Flash 1017. An experimental CLINE_CLOUD_SESSIONS=1 setting runs tasks in Cline Cloud from the TUI, and undici was raised to 7.30.1 to fix two process-crash issues.
Why it matters if you build agents
Scripts and pipelines that send one-shot prompts will now run tools without approval, so check auto-approve settings and pass --auto-approve false or --act where approval or team tools are needed. Full file reads now reach the model, which can change token use for agents that read large files. Anyone relying on the Cline provider's default model should review their saved selection and the new defaults.
Get stories like this every weekday
One short email with the day's agent tooling news.